deploy - SKILL.md Agent Skill

name: deploy description: >- Clarifies what “ship” means for kokoro-coreml: there is no Cloudflare-style multi-worker deploy script. Use for tagging releases, pushing the branch, coordinating with the macOS app repo, or verifying exports before handoff. Before treating a revision as releasable, run the repo’s primary checks (see git-commit / audit). Do not use when the user only wants local experiments with no remote or release intent.

Deploy (kokoro-coreml)

Purpose

This repository is a PyTorch → Core ML conversion and model-artifact tree for Kokoro TTS. It is not the Roast Cloudflare monorepo: there are no pnpm deploy:* scripts or Workers/Pages pipelines here.

“Deploy” in this context means release engineering appropriate to this repo:

Git: push branches, tags, or PRs (often via git-push).
Artifacts: .mlpackage bundles under coreml/, checkpoints, and export scripts consumed by a separate macOS app (e.g. TalkToMe).
Verification: run pytest (or targeted export smoke scripts) before declaring a revision ready for downstream integration.

Default interpretation

“Ship it” / “deploy” → confirm what the user means: push to GitHub, cut a tag, or hand models to the app—not a cloud deploy unless they name another system.

Pre-ship gate (recommended)

From the repository root, when a change touches Python export, Core ML, or tests:

pytest (full test pass when tests are present).
If the change is export-only, at minimum run the relevant examples/ or test_*.py scripts the guides mention.

There is no single CI workflow in .github/workflows/ for this repo yet; treat local green tests as the main gate before pushing risky changes.

Procedure (agent)

Read README.md, CLAUDE.md (PyTorch → Core ML playbook), and the relevant README/*.md guide for the subsystem (conversion, runtime, export).
Do not run Cloudflare/Roast deploy commands—they do not apply.
Run Pre-ship gate before advising “ready to ship.”
For git remote operations, follow git-push when the user wants push
- CI (if added later).

Anti-patterns

Copying Roast pnpm deploy:staging / Workers steps into this repo.
Pushing large binary or regenerated .mlpackage changes without noting what was regenerated and how to verify.

Related skills

git-push: commit, merge, push, chase CI when CI exists.
git-commit: post-commit pytest awareness (non-blocking heads-up).
audit: full review when the user says audit.